# Building a Reliable Citizen-Science Bird Survey Pipeline

Citizen-science bird surveys work best when the field protocol is simple but the data pipeline is strict. In our spring survey, volunteers submitted 10-minute point counts from parks, schoolyards, wetlands, and backyards, recording every bird seen or heard within a 50-meter radius. The goal was not to produce a perfect census, but to collect repeatable observations that could reveal migration timing, habitat use, and changes in local abundance.

## Field Data Model

Each checklist included observer ID, location, start time, duration, weather, habitat type, and a species list with counts. We required GPS coordinates from the mobile app and rounded them to five decimal places before storage, which preserved site precision while avoiding unnecessary exposure of private locations. To reduce false positives, rare species and unusually high counts were flagged for review before entering the public dataset.

```yaml
survey:
  duration_minutes: 10
  radius_meters: 50
  required_fields:
    - observer_id
    - latitude
    - longitude
    - start_time
    - species_count
```

## Quality Control

The most useful validation step was comparing each submission against an expected regional species list for the survey week. A Pine Siskin report in March might pass automatically, while the same report in late June would be sent to a reviewer with the original notes and any attached audio. We also tracked observer experience, not to reject beginner data, but to weight uncertain records appropriately during analysis.

## Analysis Notes

After cleaning, we grouped observations by habitat and week, then calculated reporting rates rather than raw totals. This helped normalize differences in volunteer effort, since one popular wetland received 80 checklists while some neighborhood sites received only three. Early results showed Tree Swallows arriving about nine days earlier at stormwater ponds than at forest-edge sites, likely because open water produced insects sooner.

## What Worked

The biggest technical lesson was that small constraints at collection time prevented large cleanup costs later. Fixed-duration counts, required coordinates, controlled habitat labels, and automated review flags made the dataset much easier to trust. Citizen-science data will always be uneven, but with a careful schema and transparent validation rules, it can still support serious ecological questions.